RAG in Azure with OpenAI and ChatGPT LLM model

Learn how Retrieval Augmented Generation enables ChatGPT to access internal enterprise data by indexing documents with embeddings, retrieving top matches via cosine similarity, and feeding them into prompts.

Explore retrieval augmented generation (rag) to enhance chat models by querying external data, using embeddings and vector search with Azure OpenAI, to deliver accurate, sourced answers.

Create and connect Azure AI resources using the portal to build a retrieval augmented generation workflow, including AI Studio deployments, a hub and project, and an Azure AI search service.

See a hands-on demonstration of provisioning azure resources via command line, including creating resource groups, ai services, ChatGPT deployments, endpoints, keys, and a YAML connection to azure ai studio.

Demonstrates retrieval-augmented generation with Azure by indexing markdown docs, embedding them with OpenAI, and retrieving chunks via Azure AI Search to answer questions using GPT-4.

Count text tokens with a utility function and understand token limits for OpenAI chat models and embedding. The demo scans Azure docs, flags files over 8191 tokens, and prepares chunks.

Demonstrates cleaning markdown content by removing links, images, and double asterisks with a Python function, preparing markdown files for LLM workflows in a RAG setup.

Create a vector embedding for input text with a function that generates the mathematical representation as a multi-dimensional vector of 3000 dimensions, with values from -1 to 1.

Chunk markdown documents into sections to fit embedding token limits, generate vector embeddings for each chunk, and save metadata in JSON for Azure AI search indexing.

Demonstrates creating and uploading vector-based content to Azure AI Search, defining an index with fields for page and chunk data, using semantic search and embeddings for retrieval.

Upload JSON chunks to Azure AI Search by looping through the output directory, loading each JSON, and uploading documents, then verify and view search results with relevance scores.

Demonstrates writing embeddings into an Azure AI search vector index and performing a vector search to retrieve the three nearest documents, showing titles, dates, and chunk content with scores.

Learn retrieval augmented generation by embedding a question, searching documents with Azure AI Search, and feeding chunks into Azure OpenAI ChatGPT to generate a response.